A HIERACHICAL APPROACH TO QUANTIFY UNCERTAINTY IN MULTI-SCALE MODELING OF RIVERINE ECOSYSTEMS AND RESPONSES OF FISH POPULATIONS

Models for large river fish populations are dependent on habitat conditions linked to hydrological variability of the river itself, which is linked to variation in weather variables, which are ultimately linked to potential climate variations. There is uncertainty in each linkage, and also in individual process models and parameters upon which the models rely. The hierarchical modeling approach we will present should help to account for these uncertainties, in particular the variability of relevant climate conditions across temporal and spatial scales, so projections of community or population response to a given climate change scenario include realistic measures of uncertainty. The approach incorporates various independent sources of observations, includes established scientific knowledge, and addresses uncertainties by linking system components together using formal rules of probability..

Physical, natural and biological sciences rely heavily on numerical models for structure and evolution of features of environment at various scales in space and time. Varying amounts of relevant data are collected at different scales and with varying levels of completeness. These models and data are fraught with uncertainty. With such uncertainty, scientists from many disciplines recognize that prediction (forecasting) of complex phenomena is statistical or stochastic by nature. For ecological modeling, Levin and others (1997) noted “…models should not be expected to predict where every tree will be at each point in time; only aggregate statistical properties can be reliably predicted, typically over broad spatial and temporal scales.” Any approach to modeling ecological phenomena should rely on information deemed relevant and produce predictive output that is responsive and “honest” with regard to intrinsic uncertainties. It should also be capable of combining information and data from diverse sources, relevant at differing scales in space and time, and of varying quality. It must also account for nonlinearities present in hypothesized models for physical and biological processes, as well as complex interactions across subsystems. The hierarchical Bayesian modeling approach offers such a paradigm for development of a hybrid deterministic stochastic downscaling model. The Bayesian approach seeks the combination of science and statistics expressed mathematically through probability distributions (for example, Amstrup and others, 2007).

This talk will focus on development of probabilistic linkages to quantify implications of climate on fish populations of the Missouri River ecosystem. This approach is a hybrid between physical (deterministic) downscaling and statistical downscaling, recognizing that there is uncertainty in both. Ultimately, the model must include linkages between climate and habitat, and between habitat and population.

Amstrup, S.C., Marcot, B.G, and Douglas, D. C., 2007, Forecasting the range-wide status of polar bears at selected times in the 21st century: USGS Administrative Report to the U.S. Fish and Wildlife Service, Reston, Virginia, p. 126.

Levin, S.A., Grenfell, B., Hastings, A., and Perelson, A.S., 1997, Mathematical and computational challenges in population biology and ecosystems science: Science, v. 275, no. 5298, p. 334-343.